Code
There are two components to the system: data processing/analysis and data collection. Code for each component is provided below.
Analysis and Baseline System
These R scripts were used to analyze the dataset and to implement a baseline QBT system. You will need to get R to run it. If you are new to R (or maybe you prefer IDEs), we suggest to also install RStudio.
Results Errata
The results for the melodic contour classification shown in Table 5 of the ISMIR'13 paper where slightly underperformed. This was due to over-thresholding differences, which mistakenly clustered together "notes" that could be differentiated by using a smaller threshold value. The corrected values are:| Accuracy (%) | |||
|---|---|---|---|
| Condition | Top 1 | Top 5 | Top 10 |
| Rhythmic contour | 51.92 | 70.82 | 78.46 |
| Melodic contour | 41.52 | 56.88 | 65.97 |
| Both contours | 55.22 | 71.68 | 79.41 |
| Significance analysis | |||
| Χ2(1, N=3,365) | 16.24 | 1.45 | 2.29 |
| p-value | 0.000058 | 0.23 | 0.13 |
Data Collection System
The dataset was collected using an iOS app, which fed a backend implemented using Ruby-On-Rails. The code can be found here.